Data model

On disk

DATA_DIR is the git working tree. Items are real files; the SQLite DB and .env live outside it so they're never committed.

DATA_DIR/
  items/
    <folder>/<slug>-<id>.<ext>      # one file per item
  assets/<token>.<ext>              # pasted images
  .gitignore
scratchpad.db                       # SQLite index — OUTSIDE DATA_DIR
.env                                # secrets — OUTSIDE DATA_DIR

Extensions encode the type: code uses its language extension (.go, .py, …), drawings .excalidraw, mindmaps .mind, docs .doc (HTML), boards .kanban (JSON). The <id> keeps filenames unique.

SQLite tables

All rebuildable from disk except shares (and the generated session secret in settings).

items(    id, title, type, path, language, folder, created_at, updated_at )
shares(   token, item_id, expires_at, created_at )   -- expires_at = 0 means never
links(    from_id, to_title )                         -- [[wiki-link]] graph
settings( key, value )                                -- session secret, etc.

Reconcile

On boot (after a git pull), Scratchpad walks DATA_DIR/items, upserts each file into items, rebuilds the links index, and prunes rows whose file is gone. So the index always reflects what's on disk — including changes pulled from another machine.